Sources We Track

Most conservative of AI lab CEOs. Track record of delivering on specific technical claims (AlphaFold). Longer timeline estimates tend to be more accurate than peers.

Academic rigor and independent perspective — not affiliated with any AI lab. Safety concerns have been directionally validated. The Safety Report represents consensus of 100+ experts, not just personal views.

Directionally correct on capability trajectory. Timelines tend 6-12 months aggressive. Strong on technical predictions, weaker on societal impact timing.

Practitioner perspective grounds his predictions. Good at near-term workflow predictions. 10-year timeline (P-010) is notably more conservative than peers — may indicate better calibration.

Strong on negative predictions (what won't work). Weaker on positive predictions (what will work instead). Bubble prediction (P-012) looks increasingly wrong — may indicate blind spot on market dynamics vs technical assessment.

Practitioner perspective with verifiable output. Working at the frontier of AI-assisted development. May overstate generalizability from his own elite-level usage to average developers. Power-law applies: his results represent the 90th+ percentile.

Revenue figure is extraordinary and verified. The 12-employee figure makes this the most compelling evidence for AI-enabled small team disruption. Not a prediction source per se — more a living proof point.

Most conservative credible estimate of AI economic impact. 1.5% GDP by 2035 is significantly below Goldman ($7T) and PwC ($15.7T). Historical accuracy on fiscal projections is strong. May underweight AI adoption speed.

Consistently optimistic on timelines. Vague language ('not-very-distant future') makes predictions hard to falsify. Directionally correct but systematically overconfident on speed.

Best calibrated on 'what won't happen' predictions. Humanoid robot skepticism (P-014) partially vindicated. Reliability/profit claim (P-015) increasingly challenged by revenue data. Tends to anchor too strongly on limitations.

Conservative and well-calibrated. 20K/month job loss estimate (P-020) is tracking well against early data. Tends to understate disruption speed but overstate near-term job losses. Good base-rate methodology.

Good at enterprise adoption tracking. Agent adoption prediction (P-021) already exceeded — may indicate they were conservative. Failure rate prediction (P-022) is well-grounded in historical data. 2035 revenue projection (P-023) is too far out to assess.

Conservative methodology grounded in surveys. 14% career change (P-024) aligns with historical base rates — may be measuring normal churn rather than AI-specific disruption. Good at order-of-magnitude estimates.

Demonstrating real disruption in legal — not just a prototype. $195M ARR with $11B valuation shows market conviction. Revenue-to-valuation ratio suggests high growth expectations. Incentive to overstate AI's legal capabilities.

Gartner predictions tend toward the dramatic for marketing impact but are grounded in enterprise survey data. Their AI predictions have been directionally correct. The facial verification prediction is specific and measurable.

Deloitte's financial services research is well-sourced and methodologically sound. The $40B fraud figure is aggressive but backed by current trajectory data (deepfake files 16x since 2023, 40% of biometric fraud).

Academic researcher with deep technical expertise in deepfake generation and detection. His warning that '2026 is the year you get fooled' is opinion but backed by detection accuracy data (human accuracy only 24.5% on high-quality deepfakes).

Institutional voice — projections tend toward moderate scenarios. The 170M jobs created by 2030 is optimistic and assumes effective policy response. Discount by standard institutional optimism bias.

Aggressive timeline predictions but backs them with massive capital commitment. Llama open-source bet was contrarian and proved right. May overstate internal AI adoption timelines for PR value.

Conservative institutional research. The 1% GDP contribution figure is a hard number verifiable against BEA data. Useful as cross-check against Goldman Sachs projections.

Most aggressive forecaster in our database. 'Fully automated in 18 months' (P-016) is an extreme outlier. Has product-selling incentive that may inflate predictions. Discount timelines by 2-3x.

Early stage — high valuation but limited revenue data publicly available. The consulting disruption thesis is compelling but unproven at scale. Palantir background gives credibility on enterprise deployment.

Impressive technical achievement (full autonomous tax return). But tax preparation is highly structured — success may not generalize to less structured accounting work. Revenue data limited.

Practitioner platform with large-scale engineering telemetry. The negative findings (incidents up, failure rates up) lend credibility — not marketing-driven data. Limited to companies using their platform, which skews toward larger engineering orgs.

HR platform with commercial interest, but survey methodology is transparent and sample size (500+) is reasonable. Key finding — 88% investing but only 11% integrated — is consistent with Gartner and McKinsey data.

Directionally correct on trends, systematically overconfident on magnitudes and timelines. CapEx projection (P-017) is aggressive but within historical range. Inference cost claim (P-018) was factual, not predictive. Discount specific numbers by 30-40%.

Bold action (40% workforce cut) but Bloomberg/HBR flag possible 'AI washing' — some cuts may be restructuring unrelated to AI capability. Track record as operator is strong, but claims about AI replacing workers may be partially performative.